Writing study: One for Two: A Unified Framework for Imbalanced Graph Classification via Dynamic Balanced Prototype

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Guanjun Wang, Binwu Wang, Jiaming Ma, Zhengyang Zhou, Pengkun Wang, Xu Wang, Yang Wang · Official ICLR 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Learning & optimization algorithms (primary) · Architecture & representation design.

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

38 physical pages

Before references / appendix

≈ 10.43 page-equivalents

Main figures / tables

6 / 5

Attached appendix figures / tables

10 / 19

Attached appendix span

≈ 21.87 page-equivalents

Before references / appendix: ≈ 27.45%References / intervening material: ≈ 15%Attached appendix: ≈ 57.55%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1 INTRODUCTION	1	≈ 1.06 pp.	2.79%
2 RELATED WORK	2	≈ 0.84 pp.	2.21%
3 PRELIMINARIES	3	≈ 0.16 pp.	0.42%
4 METHODOLOGY	3	≈ 3.21 pp.	8.45%
5 EXPERIMENTS	6	≈ 3.94 pp.	10.37%
6 CONCLUSION	10	≈ 0.32 pp.	0.84%
ETHICS STATEMENT	11	≈ 0.11 pp.	0.29%
REPRODUCIBILITY STATEMENT	11	≈ 0.21 pp.	0.55%
REFERENCES	11	≈ 5.71 pp.	15.03%
Appendix	17	≈ 0.97 pp.	2.55%
A USE OF LLM	18	≈ 0.09 pp.	0.24%
B RELATED WORK	18	≈ 1.38 pp.	3.63%
C METHOD	19	≈ 0.21 pp.	0.55%
D THEORY PROOF	19	≈ 5.08 pp.	13.37%
E ALGORITHM PSEUDOCODE	24	≈ 0.72 pp.	1.89%
F DEFINITION OF GRAPH CLASS IMBALANCE AND TOPOLOGICAL	25	≈ 0.81 pp.	2.13%
G DATASET	26	≈ 4.07 pp.	10.71%
H EXPERIMENT SETTING	30	≈ 1.04 pp.	2.74%
I ADDITIONAL EXPERIMENTAL RESULTS	31	≈ 7.03 pp.	18.5%
J LIMITATION DISCUSSION AND FUTURE WORK	38	≈ 0.46 pp.	1.21%

All detected figure and table locations

Main text

• Figure 1 · p. 3
• Table 1 · p. 6
• Table 2 · p. 7
• Table 3 · p. 8
• Table 4 · p. 9
• Figure 2 · p. 9
• Figure 3 · p. 9
• Figure 4 · p. 9
• Figure 5 · p. 10
• Figure 6 · p. 10
• Table 5 · p. 10

Attached appendix

• Table 6 · p. 24
• Table 7 · p. 25
• Table 8 · p. 26
• Figure 7 · p. 27
• Table 9 · p. 28
• Table 10 · p. 29
• Table 11 · p. 30
• Table 12 · p. 30
• Table 13 · p. 30
• Table 14 · p. 32
• Table 15 · p. 32
• Table 16 · p. 33
• Table 17 · p. 33
• Table 18 · p. 34
• Table 19 · p. 34
• Table 20 · p. 34
• Table 21 · p. 34
• Figure 8 · p. 35
• Figure 9 · p. 35
• Figure 10 · p. 35
• Figure 11 · p. 36
• Figure 12 · p. 36
• Figure 13 · p. 37
• Figure 14 · p. 37
• Table 22 · p. 37
• Table 23 · p. 37
• Table 24 · p. 38
• Figure 15 · p. 38
• Figure 16 · p. 38

Argument and organization

1. Introduction and preliminaries

   Distinguish class-frequency imbalance from graph-size imbalance before proposing a shared remedy.

2. Method and experiments

   Build topology encoding, perturbation and balanced prototypes, then evaluate isolated and intertwined imbalance.

3. Appendices A–J

   Separate related work, derivations, construction protocols, extended tests and limitations.

Problem, claim and contribution

Two clearly named imbalance axes create a testable unification claim; graph-level classification is distinguished from node- and edge-level tasks.

Sections 1–3; Appendix B and F. · Main paper

Argument and section order

Main experiments separate class, topology, intertwined imbalance, ablations and 3D conformations. This organization is more informative than one aggregate leaderboard.

Sections 5.1–5.8, pages 7–10. · Main paper

Figures, captions and visual explanation

Figure 1 expands perception and balancing beneath the pipeline, a useful overview-to-detail pattern. Dense arrows, textured backgrounds and small mathematical labels reduce legibility at normal page size.

Figure 1, page 3. · Main paper

Tables, comparisons and evidence

Backbone-with/without pairs make the claimed plug-in contribution directly inspectable. The AirGraph bubble plot places accuracy, time and memory together, whereas radial bars make exact cross-dataset comparisons harder.

Table 4 and Figures 2–4, page 9. · Main paper

Appendix and reproducibility

Dataset construction tables preserve different binary, multiclass and topology splits. AirGraph's forecasting description would benefit from an explicit chronological partition and leakage protocol.

Appendices G.1.2–G.1.5, pages 26–30. · Main paper

Clarity and writing conventions

Theory should separate a variational upper bound and local Taylor approximation from exact equivalence. Bias placement after TopK and the activation-count update deserve implementation-level clarification before reuse.

Equations 20–35 and Algorithm 1, pages 20–25. · Main paper

What to reuse—and what not to infer

Transferable lesson: Organize a unification paper around independent failure axes, then test each axis and their intersection with clearly paired controls.

Boundary: Balanced prototype use alone does not establish class fairness; the theoretical reduction and discrete routing update are not independently certified. Low-imbalance results include exceptions to broad superiority wording.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–11: Sections 1–6, ethics and reproducibility; references 11–16.

Appendix / supplement

Pages 17–38: contents and Appendices A–J, including proofs, pseudocode, splits, all supplementary tables and limitations.

Visually inspected PDF pages

3, 9

• ICLR 2026 final proceedings · 38 pages